Comparative Study On The Prediction Of Under Five Mortality Rate In China Based On Machine Learning Model

2020 is the decisive year for China to build a well-off society in an all-round way,and the level of children's health related to the future of the nation,it has also attracted more and more attention from all sectors of the society.In 2019,the National Health Committee held a conference on the development of China's maternal and children's health.According to the conference,the neonatal mortality rate,infant mortality rate and under 5-year-old mortality rate decreased by more than 85% compared with 1991.At the same time,the under 5-year-old mortality rate is also an important indicator for our government to manage and make decisions on maternal and children's work.The data in this paper mainly comes from the statistics yearbook of 2019,published under 5-year-old mortality rate in monitoring areas of 1991-2018,including the neonatal mortality rate(national,rural,urban),infant mortality rate(national,rural,urban)and the under 5-year-old mortality rate(national,urban,rural).First of all,this paper makes a descriptive statistical analysis of the time series of mortality,and draws the conclusion that the indicators of children's health development in each region will continue to show a downward trend through Cochran-Armitage trend test.In the past,the development of children's health indicators in China was only a simple extrapolation of historical data,so sometimes there are unreasonable goals.Although some scholars have used a single prediction model to predict the under 5 mortality rate,there is no horizontal comprehensive comparison between traditional time series prediction model and machine learning model.In order to reflect the impact of many factors on the health development indicators of children under five years old in China,this paper adopts the idea of hierarchical prediction,that is,to predict the areas with different economic and social development levels in urban and rural areas.Using GM(1,1),radial basis function neural network model and support vector regression model to predict the under 5 mortality rate,the results of each model are comprehensively evaluated from the back generation fitting effect and point prediction effect,and the prediction effect of machine learning model is compared with that of grey model horizontally,so as to provide a reasonable target for the development of children in China for reference in methodology and theory.The results show that the prediction effect of machine learning model is better than that of traditional time series prediction model for the prediction of time series data of children's health development indicators in this paper.What's more,the prediction effect of support vector regression model is the best in most cases,which can reduce over fitting from the perspective of model itself and has better generalization ability.The prediction results show that in 2020,China's under five children's health development indicators will still show a downward trend,but at the same time,the gap between urban and rural areas is still obvious,and the gap between urban and rural income and medical and health conditions is one of the reasons,so improving the health level of children in rural areas is the focus of the program.The innovation of this paper lies in the introduction of the support vector regression model to predict the national under five mortality rate,thus proving the universality of the support vector regression model for the short-term time series' prediction,such as under five mortality rate,expanding the application field of the SVR model.And the prediction effect of the machine learning model and the traditional time series prediction such as grey model is compared from the aspects of model's robustness and the prediction model's accuracy.